socio-economic determinants of student mobility and...
TRANSCRIPT
Working Paper Series
Department of Economics
University of Verona
Socio-Economic Determinants of Student Mobility andInequality of Access to Higher Education in Italy
Umut Turk
WP Number: 8 May 2017
ISSN: 2036-2919 (paper), 2036-4679 (online)
Socio-Economic Determinants of Student
Mobility and Inequality of Access to Higher
Education in Italy∗
Umut TURK †
Abstract
This paper introduces a modified version of the Hansen-gravity model as a
framework to estimate the accessibility of higher education (HE) institutions
in Italy from equal opportunities perspective. The key assumption underly-
ing gravity models is that accessibility decreases with spatial distance from
opportunities. The paper extends the gravity equation to allow for the inclu-
sion of socio-economic factors influencing the access to HE. The findings reveal
differences in response to quality and to other institutional characteristics by
parental background and gender. Finally, decomposition of overall inequal-
ity into spatial and aspatial components reveals both the physical and social
distance between groups of students seeking higher education opportunities in
the country.
Keywords Spatial Interaction, Higher Education Accessibility, Gravity Model,
Equality of Opportunity
∗This paper has benefited from the insightful comments and suggestions of the partici-
pants of 9th GRASS Workshop in Lucca, Italy; RSAI/ERSA Workshop on Regional Science
and Urban Economics Spatial Perspective of Human Capital in Barcelona, Spain; Joint
meeting NECTAR Cluster 6 Accessibility/Regional Science Academy in Paris, France.†Department of Economics, University of Verona, Verona, ITALY. Email address:
1
1 Introduction
An intuitive way to increase spatial accessibility is to decentralize the service
in question. This was the strategy implemented by the Italian authorities in
the period 1990-1998. With one of the lowest participation and graduation
rates in Europe, the supply of higher education (HE) was expanded drasti-
cally. The reforms required larger universities to set up new types of faculties
and 9 new higher education institutions were established as a result of the
decentralization process (MIUR). However these reforms took place without
any field examination of accessibility or demand (Bratti et al., 2008). More
than a decade later, there is no explicit measure of spatial accessibility of the
universities in the country.
In a system granting free access to HE for every potential candidate,
the foremost aim of policy makers is to guarantee full accessibility to the
service irrespective of the location of residence. Previous research has focused
on measuring accessibility through an examination of the match between the
locational distribution of facilities or services and the locational distribution
of residents (Talen & Anselin, 1998). In this framework, the spatial distance
between residence of origin and the location where opportunities are located
is regarded as an important factor determining the spatial accessibility. The
underlying idea is that people from more isolated locations face larger costs to
access to opportunities, with costs growing with spatial distance.
Since the residential location of students prior to HE enrolment is
detemined by parents, inequalities in access to HE across students’ locations
of origin should be regarded as unfair. Modern theory of inequality, building
on equality of opportunity (EOp) arguments, suggests that the differences in
outcomes due to factors that are beyond individual responsibility are unfair
and should be compensated by society (Dardanoni et al., 2006). Reducing
geographical disparities in accessibility can be seen, in fact, as a way of leveling
the playing field (Roemer, 1998) and providing equal opportunities to benefit
2
from HE irrespective of the place of origin. The geographical location can be
an unfair source of accessibility on the large scale: a student living in an urban
area where HE services are supplied faces smaller costs of transportation, lower
opportunity costs in commuting and no housing costs, compared to students
living in the countryside who need to commute or move to benefit from HE
services. However, focusing only on geography may leave the influence of
socio-economic factors, in relation to gender, experiences at home and parental
background, unexplored.
The paper argues the existence of a gradient of economic circum-
stances of origin on distance elasticity. Although distance matters in explain-
ing accessibility, there are other variables that determine differences in costs of
movement, correlated with distance which, at the same time, might influence
the distribution of accessibility. This paper tries to single out the contribution
of spatial distance and economic circumstances on inequality in accessibility
to HE by using a multidisciplinary approach, where the problem of disparities
in spatial access is redefined on the basis of both the physical distance from
universities and the social distance between student groups that generates an
additional inequality in access within the same location. The paper looks at
the variability in access both when focusing on comparisons of people located
at different origin points from HE, but all sharing the same family background
(highlighting the share of inequality due to spatial distribution of HE institu-
tions in the country) and when comparing people located at the same origin
points but with differing backgrounds of family origin, which is taken as a
proxy of the ability of families to cover costs of displacement and, if possible,
to compensate for distance from the location of origin. The latter shows the
share of inequality in access due to the socio-economic background of students.
In an empirical application, the paper sequentially employs a model
and an index to measure overall inequality in access, which is then decom-
posed into its geographical and socio-economic components ; first a spatial
interaction model (SIM) is used to disentangle the migration dynamics of dif-
3
ferent student groups. Being flexible and simple enough, these models enable
the investigation flows between origins and destinations (Sen & Smith, 2012).
Actual flows of commodities, information, emails, phone calls, money and of
people along with any other sort of movements are likewise applicable to SIMs
(see Haynes & Fotheringham, 1984; Sen & Smith, 2012, for reviews). In the
present application, student flows from parental residents to universities are
defined as interactions between localities and, to account for socio-economic
factors in place, the total observed flows of students are partitioned into sub-
groups each representing a different type. It is a common practice for EOp
studies to partition the population according to exogenous factors, which are
assumed to be beyond people’s control (see for instance Checchi & Peragine,
2010; Ferreira & Gignoux, 2011) and resulting subgroups are defined as types
(Roemer, 1998). For the second step, the parameters that are distinctively
calibrated for each type by the SIMs are imported to a Hansen (1959)-like
index to measure potential accessibility for 110 Italian provinces (NUTS3 level
regions). Finally the inequality in accessibility among provinces is decomposed
as follows: the access score in each province is replaced with its average access
score across socio-economic groups hence only variation is allowed to be due to
the geographical distribution of universities. Then the access scores computed
for each socio-economic group is replaced with its average access score across
provinces hence remaining variation is allowed to be due to socio-economic
backgrounds. This operation enables investigating the relative contributions
of spatial and aspatial factors.
The paper contributes to the literature by extending classical spa-
tial accessibility analysis to incorporate the socio-economic circumstances of
students in a spatial accessibility measure for Italian HE institutions. This
practice goes beyond the mere concern of inequalities on outcomes. For the
spatial accessibility analysis this means that the inquiry may shift from ”spa-
tial accessibility where?” to ”spatial accessibility where and for whom?”. It
also contributes to the EOp literature by showing how the spatial dimensions
4
of the theory can be incorporated into models that rely solely on geography.
Finally, the findings in this paper provide highly detailed information for policy
makers regarding which groups of students to target and specifically in which
locations the assistance is needed most.
The remainder of the paper is organized as follows: the second section
introduces the model and the accessibility index adopted, the third section sets
out the data and variables, the fourth section shows empirical method for cali-
bration and findings where inequality in access is decomposed into within and
between components. Finally the conclusions and possible policy implications
are given in the fifth section.
2 Theoretical Framework
This section presents the model adopted for student flows and the potential ac-
cessibility index. The link between these two builds on the distance parameter
assumed to reflect both physical and social costs in migrating or commuting to
destination universities. The response to distance is expected to be conditional
on the socio-economic background of students.
2.1 A Spatial Interaction Model of Student Mobility
Spatial interaction models are used to predict the size of spatial flows between
origins and destinations in areas of interest. They have been used mainly
for transportation and environmental planning, then developed further for a
variety of applications where a movement and/or interaction takes place. Re-
cently, most applications relate to health system planning, decisions concerning
hospital locations, the analysis of interaction between patients and physicians
as well as labour studies such as job accessibility and an investigation of the
daily commute to work (Mayhew et al., 1986; R. M. Wilson & Gibberd, 1990;
5
Reggiani et al., 2011).
Particularly with regard to SIMs applied to HE choice, Sa et al. (2004)
studied the demand for HE in the Netherlands, given the attractiveness and
accessibility of universities. Alm & Winters (2009) correlated the distance
from parental residence to state HE institutions with tuition, financial aid and
school quality as institution fixed characteristics and found a varying deter-
rence effect of distance in relation to these characteristics. Cooke & Boyle
(2011) included several origin and destination attributes to SIMs, including
the number of high school graduates in origins, employment growth both in
origins and destinations and the relative quality of amenities. Singleton et
al. (2012) integrated SIMs with geodemographic analysis, and looked at both
socio-spatial conditions in the neighbourhood and the attractiveness of des-
tinations. For the Italian data, Dotti et al. (2013) investigated the role of
universities in attracting successful students to certain regions and to settle
down there after graduation. These studies estimate the distance elasticity
of university choice given the attributes of origins/destinations and some also
include university characteristics. However they do not incorporate the socio-
economic profile of students in the analysis. The present paper examines the
role of socio-economic characteristics of students in distance elasticity. Fur-
thermore, based on the distance elasticity values, the paper transforms SIMs
into an explicit measure of accessibility.
SIMs can incorporate a range of origin and destination constraints
and take a number of forms according to this constraint structure. The fol-
lowing formula is a production-constrained form of SIMs that suggests that
the interaction between any two units must be directly proportional to the
masses of origin and destination and inversely related to the distance between
them. The basic assumption is that a positive interaction between each pair
of locations exists 1.
1(see A. S. Fotheringham & Webber, 1980; A. Fotheringham & O’Kelly, 1989; Sen &
Smith, 2012, for reviews)
6
Tij = KiOθiD
αj f(dij) (1)
Tij refers to student flow from their residence i to university j
Oi origin dummies for 110 Italian provinces (NUTS3 level regions)
Dj total number of students reaching university j
Ki = [∑
j Dαj f(dij)]
−1 is the balancing factor ensuring that the marginal
total constraint∑
j Tij = Oi is satisfied.2
f(dij) = d−βij where dij is the Euclidean distance between city i and
university j
For this application, the model is extended to include several univer-
sity fixed characteristics and interactions with distance. Finally the following
model is obtained:
Tij = KiOθiD
αj Sj
γLjη exp(−βln(dij) + µδij +
∑
l
λlln(dij)Ujl) (2)
where Sj and Lj are two variables accounting for the attractiveness of a des-
tination. Following the previous studies (see for example Lowe & Sen, 1996;
Gitlesen & Thorsen, 2000; McArthur et al., 2011) a Kronecker delta is added
to the model as follows:
δij =
1 i = j
0 otherwise
2 With Ki the model becomes production-constrained. The choice of this model is jus-
tified by the fact that most of the programmes are provided in an open-access fashion in
Italy. Therefore, theoretically, students are free to choose any destination desired hence the
model is not constained by destination (not attraction constrained) but to make sure that
the number of trips produced by an origin do not exceed the number of residents, the model
is constrained from production side. For the formal development see A. G. Wilson (1971)
7
The common interpretation of µ is that it reflects the benefit of residing and
studying in the same city, or a start-up cost in case i and j are not in the same
province. Furthermore f(dij) interacts with several destination characteristics
Ujl where l is the number of interaction terms and λl is the distance elasticities
given the institutional characteristics.
2.2 Adopted Accessibility Index
In this paper, the accessibility concept is interpreted as the potential availabil-
ity of HE given the spatial distribution of institutions in the country. The roots
of the index go back to Hansen (1959) when he first proposed the following
gravity model of accessibility:
Ai =J∑
j=1
Sjd−βij
where Ai is a measure of accessibility , Sj is the number of opportunities at
the destination and dij is the distance between an origin and a destination. A
similar accessibility index can be constructed as follows:
Ai =∑
j=1
Cjdβij
δij(3)
where
δij =
exp(µ) i 6= j
1 otherwise
µ and β are the two parameters that channel (2) to (3) and are
calibrated beforehand by the production-constrained SIM (2). Cj is the total
number of places offered by each institution. Additionally, the index discounts
accessibility when i and j are not located in the same province by δij.
8
3 Data and Variables
Table 1 shows the variables used in the analyses carried out in this paper. The
data is extracted from a data survey (Inserimento professionale dei laureati,
2011) including 14,000 male and 17,400 female graduates in 2007 and data from
MIUR (Ministry of Education, 2003-2004-2005). The survey data includes the
information of student residence in 110 Italian provinces (NUTS3 level regions)
before enrolling to a university and the name of university enrolled. The actual
flow of students between the province of residence and the exact addresses of
universities is extracted and stacked into a table as a column vector as the
variable of interest.
3.1 Types
This paper argues that at least three aspatial factors are particularly relevant
to the study of HE accessibility. Firstly, the role of parental education is a well-
explored factor that affects the educational choices and outcomes of students.
Specifically in the Italian context, the educational level of parents is found to
be highly influential for the academic attainment of Italian students (Checchi
et al., 2003; Bratti et al., 2008). Higher HE participation rates and less drop-
outs are observed for students with highly educated parents(Checchi & Flabbi,
2007; Brunori et al., 2012). Moreover, since commuting or migrating to a place
involves a cost, the financial condition of families is another aspatial factor
relevant to access (see Frenette, 2003; Lupi & Ordine, 2009). Finally, even
though education is the primary area where women have made substantial
gains and now largely out-perform men (DiPrete & Buchmann, 2006), the
question whether there are systematic differences in spatial access to education
among males and females remains an important one.
Observed flows are partitioned according to three sets of proxies re-
ferring to the socio-economic circumstances of students. Each subgroup forms
9
a type, which cannot be chosen by students. Accordingly, three circumstance
variables are employed as shown in Table 1: the presence of at least one highly
educated parent at home where the alternative is both parents with basic ed-
ucation. Here “basic education” means the 8 years of compulsory schooling
in Italy. ”High education” consists of parents with at least a bachelor’s de-
gree. In the survey 40.66% of mothers and 39.69% fathers are categorized as
basically educated , and 14.56% and 20.06% as highly educated, respectively.
Survey data contains information on parents’ professions. This information is
categorized as high and low for both fathers and mothers. Hence three groups
are constructed as both-low, both-high and one-high-one-low. The gender of
students is included in addition to the parental background. The combination
of these three categories resulted in 12 types as shown in Table 1.
3.2 Distance
The distance from parental residence to HE institutions strongly influences
the likelihood of participation and the HE outcomes of students (see for exam-
ple Gossman et al., 1967; McHugh & Morgan, 1984; Tinto, 1973; Ordovensky,
1995; Gibbons & Vignoles, 2009; Suhonen, 2014). The costs of commuting or
migrating may deter access or may impose a barrier when enrolling at a uni-
versity. For Italy, the empirical evidence confirms that geographical proximity
strongly influences the choice of university (Pigini et al., 2013). Indeed, in the
survey data, 59.28% studied in their hometown, 40.64% of students was moti-
vated by the closeness of the institution and only 9.74% by the prestige of the
university. For student mobility, distance does not only represent costs but is
also a predictor of how far students are allowed to live away from their families,
which is very relevant in Italy as it is a country characterized by strong family
ties (Alesina & Giuliano, 2010).
In this application dij is the Euclidean distance between the centroid
of city i and the exact address of university j. In the QGIS environment the
10
coordinates of city centres are matched with the coordinates of exact location
of universities and the Euclidean distance is calculated for each pair. According
to the interest of the investigation and the data behaviour it is possible to find
the exponential form, exponential square root, the log of distance or a relevant
combination of these (De Vries et al., 2009). To choose the most relevant form
of deterrence function, the predicted values are examined against observed
flows and the power specification of the distance proved most suitable for the
data 3
3.3 University Attractiveness
As far as institutonal attractiveness is concerned, previous studies of SIMs pro-
vide mixed findings. Sa et al. (2004) use university rankings as a quality indi-
cator for Dutch students, but the coefficient proves to be negative. Although
the authors explain this counterintuitive result as consumption behavior by
students in relation to HE (Sa et al., 2004), this is not entirely convincing.
Similarly Singleton et al. (2012) employ Times Good University Guide rank-
ings but set an arbitrary power of 0.5 rather than empirical derivation. Dotti
et al. (2013) construct an index identifying a province as attractive if inflows
exceed outflows, neglecting institutional attractiveness. This paper employs
two university fixed characteristics in order to account for attractiveness: the
share of successful students in the year before our sample’s enrolment and the
share of limited places provided by each university. In Italy after secondary
school, students take a national exam (Esame di Maturita ). The share of stu-
dents with the highest grades (90-100) from this test in the period 2002-2003
is included in the model (source: MIUR, 2004). Although many programmes
are offered on a free-access principle, some require entrance tests, indicating
excess demand for these programs. The proportion of limited places to the to-
3Also in previous studies the power- decay function has been found to be more suitable for
long distance interactions owing to the log-cost perception (A. S. Fotheringham & Webber,
1980; Reggiani et al., 2011).
11
tal number of places available at University is then used as a quality indicator
for the same period (source MIUR, 2004).
3.4 Interactions
Finally, several destination characteristics are interacted with distance to see
how the willingness to migrate to further destinations varies among different
types (see Gibbons & Vignoles, 2009, for a similar application to British stu-
dents).
Table 1 shows the interacted institutional characteristics as follows:
whether the university at destination is a private institution, dummies for
south, central and island locations and a dummy with value 1 for polytechnic
universities.
4 Empirical Method and Findings
The examination of the model is operated through related statistical log linear
models which were developed alongside entropy maximizing models. There
are several ways of handling spatial interaction models (see A. G. Wilson,
1971; Yun & Sen, 1994; LeSage & Pace, 2008). This study makes use of
the Poisson gravity models, 4 with the same statistical properties, producing
identical estimations to entropy maximization models (Baxter, 1982).
The Poisson gravity model takes the following form:
E(Nij) = Tij = OθiD
αj f(dij) (4)
where Nij indicates observed flows, whereas Tij is the expectation of observed
flows, treated as a random variable and assumed to have a Poisson distribution
4(see Flowerdew & Aitkin, 1982; Smith, 1987, for theoretical development)
12
(Baxter, 1982) 5.
The model is calibrated by the generalized linear model (GLM) pack-
age in R 6, where flows follow a Poisson distribution with a logarithmic link
between variables. The estimations are carried out separately for 12 subgroups.
Finally, the Poisson regression is:
Tijk = exp[constant+Oi+αDjk+µδij + βln(dij)+γSj+ηLj+∑
l
λlln(dij)Ujl]
(5)
where k = 1, 2, 3, ...., 12 represents types, Sj the share of successful students
and Lj the share of limited places at j, and Ujl a set of interactions on distance.
This regression produces an exponential value of factor for origin i and is
proportionally equivalent to the product KiOi, and is therefore equivalent to
the production-constrained model of the entropy-maximizing system.
Table 2 and Table 3 show the results of the first set of regressions
where the model has been applied to 10 groups. Groups 5 and 6 are not taken
into consideration due to the lower number of observations. As expected, dis-
tance has a very strong significantly negative effect, indicating a deterrence
impact for each group. For 1 meter decrease in the distance the expectation
number of student flows increases by factors varying form 1.499 to 1.709. The
impact is higher for female students than for male students except for those
with at least one highly educated parent. As a student’s family background
becomes more favourable in terms of the proxies specified above, the differ-
ence between male and female shrinks and ultimately female students feel less
deterred. As in previous studies, δ is significant at the 0.01 level for all groups
and positive in sign, capturing the benefit of residing and studying in the same
city. Similar values are observed for different types but with different motiva-
tions. For socially advantaged groups the parameter µ captures the fact that
these students usually live in big cities where large universities are located and
hence they do not need to migrate. On the other hand, groups 1 to 4 it may
5The probability mass function of flows is given by Pr(Tij) =exp
−Nij NTijij
Tij !6(see Dennett, 2012, for details)
13
reflect the actual startup costs where these students decide to migrate. As
far as the attractiveness of universities is concerned, Sj (the share of success-
ful students at university) is significant and positively affects flows only for
students who have at least one highly educated parent. Other students seem
to be unaffected by institutional quality. The effect is observed for Lj (the
share of limited places offered by universities) again for socially advantaged
groups. Among students with disadvantaged parental backgrounds, only fe-
male students (groups 2 and 4) are attracted to these limited positions. This
is probably due to the fact that female students are interested in faculties
such as medicine and nursing, requiring entrance tests. Hence, for students
with a poor parental background, what seems to matter is obtaining a degree
irrespective of the prestige of the University (Triventi & Trivellato, 2009).
The remaining results allow for interactions between institutional
characteristics and distance. Private universites at destination increase the
tendency of travelling longer distances for all groups. It is an expected result
since for any type deciding to enrol a private university, distance must be be-
coming irrelevant. Looking at the significance levels, polytechnics do not seem
to induce students to travel far except for groups 2 and 7. In contrast to Dotti
et al. (2013), interacting distance with macro regions, where universities are
located, shows that the central region attracts more students than the south
for all students except types 1 and 3. These two types comprise male students
with poor family backgrounds who may prefer Universities in the south due to
the lower cost of living. Finally universities located in Sicily and Sardinia fail
to attract students from all backgrounds.
Table 2-Table 3 [About Here]
After obtaining the parameter values from (5), accessibility scores are
calculated through Equation 3 for each group with their respective impedance
functions as follows:
Aik =∑
j=1
Cjdβk
ij
δijk(6)
14
where k = 1, 2, 3, ...., 12. As it for the production constrained SIM,
dij is constructed from city centroids to the exact addresses of universities (to
the largest campuses),so there is no zero distance, which in return accounts
for the self-potential (local demand) of universities within provinces. The
measured scores indicate potential access in terms of the places offered to
students. Higher scores indicate better accessibility to the 77 total number of
universities located in 101 different provinces. Maps 1-2 illustrate the access
scores of 10 groups, where darker blue indicates a higher score.
Figure 1-2 [About Here]
The first thing to note from the figures is that if a student belongs
to a socially advantaged group, then their access is relatively higher where
ever they live, except very far south in the country. Similarly for socially
disadvantaged groups, even if they live in a big city where large universities
are located, access remains low, particularly in the south. The lowest access
is observed for group 2 where the type comprising female students with lower-
class parents with a basic education. The types constructed for this paper
seem particularly relevant for female students. Access increases on average
101% from group 2 to group 12 whereas parental background does not seem
to affect male student access to HE as much as female students. From the
least to the highest, access increases 10% on average. Moreover a degree of
gender discrimination in access is observed in the first 4 groups, but lessens as
parental education and financial condition improve.
Decomposition of Access Inequality
To disentangle the relative contributions of spatial and aspatial fac-
tors to inequality, a decomposable inequality index is used 7. As shown in
Table 4, the resulting inequality is a sum of within and between inequalities.
7Mean Logarithmic Deviation is a path-independent decomposable inequality measure
(Foster & Shneyerov, 2000). It is defined as: MLD(X) = 1N
∑N
1 ln µx
xiwhere X is a distri-
bution, N population size and µX is mean.
15
The first row of Table 4 shows the inequality decomposition where the variation
within types of students is suppressed by substituting each type’s accessibility
score with its arithmetic mean. By this method the inequality between the
types of students is computed as 0.01776 and represents the contribution of
socio-economic factors to total inequality (5% of total inequality). Whereas
in the second row the variation within provinces is eliminated by substituting
each province’s accessibility score with its arithmetic mean, in this approach
the inequality between provinces is computed as 0.34637 and the contribution
of socio-economic factors to total inequality is measures at 7%.
For the sake of a better understanding of the computed inequality,
take a female student with a poor family background (group 2) living in Mat-
era, in order for her to have as much access as a male student with the same
family origin (group 1) living in the same city, she has to travel 151 km to
the nearest city, Foggia (social distance). Moreover, in order for her to have
similar access to a female student with better family origin (group12) living in
Napoli, she has to travel 460 km to the nearest city (social+physical distance).
Therefore, the findings indicate that despite the expansion of HE supply in
the country, access to HE is strongly unequal due to the spatial distribution
of opportunities with additional disparity due to socio-economic factors at the
locations of origin.
Table 4 [About Here]
5 Conclusions
This paper provides empirical evidence for the dynamics of student mobility
in Italy and measures inequality in access to HE institutions with particular
attention to the socio-economic background of students. Using a spatial inter-
action model, the flows of students to universities are defined as interactions
between provinces in Italy. The results demonstrate that poor family back-
16
ground students are impervious to university-quality effects and university
quality becomes relevant only for students with better family backgrounds.
The model allowed for interactions between institutional characteristics and
distance to see how elasticity with respect to spatial distance varies given the
heterogeneity of the universities. The results indicates that private universi-
ties attract students and increase their willingness to travel longer distances.
Universities located in the central region attract more students than in the
south and the location in Sicily or Sardinia deters flows.
As far as distance is concerned, the values of these parameters are
highly significant and negative in sign, indicating a deterrence effect for each
group of students. In line with previous studies, δ was significant at the 0.01
level for all groups and positive in sign, capturing the benefit of residing and
studying in the same city. For the second step computed deterrence functions
and δs were imported into a Hansen-like accessibility index and accessibil-
ity scores of 110 Italian provinces to 77 Italian universities were computed.
The results show that socio-economic background matters especially for fe-
male student mobility. Finally, the share of aspatial factors in inequality of
access between types proved to be 5% with the first approach and 7% when
computed with the second approach.
This paper contributes to the accessibility literature by a multidis-
ciplinary approach providing a spatial accessibility measure for Italian HE
institutions with particular attention to socio-economic sources of inequality.
It also contributes to EOp literature by showing how spatial dimensions of
EOp could be incorporated into models that rely solely on spatial elements.
Furthermore, the investigation of 10 types provides a clear ranking. In other
words, through this application, this study empirically shows which socio-
economic group is better off and by how much. Finally, this study is the first
attempt to define parental location, clearly an exogenous factor for students,
as a circumstance.
17
The findings provide highly detailed information for policy implica-
tions. In order to increase accessibility three policy strategies can be adopted.
Firstly, an effective policy may target the types with lower potential acces-
sibility to assist them though loans, scholarships and grants. Secondly, the
geographical locations where accessibility is lower can be identified and ac-
cessibility can be increased by the reduction of geographic barriers for cities
such as Nuoro, Brindisi, Ragusa and Belluno where new universities and/or
places may be set up. Finally a combination of these two can be used. For
instance, the empirical evidence in this paper shows that female students with
disadvantaged family backgrounds located in southern Italy would benefit the
most from HE funding. More precisely the identification of inequality resulting
from gender and geography can be extracted from the findings as follows: for
a female student living in the south with low income parents both with a basic
education, on average the potential accessibility is 146.15% lower than a male
student living in the North with better family origin. These examples can be
extended to determine a variety of policy strategies.
18
References
Alesina, A., & Giuliano, P. (2010). The power of the family. Journal of
Economic growth, 15 (2), 93–125.
Alm, J., & Winters, J. V. (2009). Distance and intrastate college student
migration. Economics of Education Review , 28 (6), 728–738.
Baxter, M. (1982). Similarities in methods of estimating spatial interaction
models. Geographical Analysis , 14 (3), 267–272.
Bratti, M., Checchi, D., & De Blasio, G. (2008). Does the expansion of higher
education increase the equality of educational opportunities? evidence from
italy. Labour , 22 (s1), 53–88.
Brunori, P., Peragine, V., & Serlenga, L. (2012). Fairness in education: The
italian university before and after the reform. Economics of Education Re-
view , 31 (5), 764–777.
Checchi, D., & Flabbi, L. (2007). Intergenerational mobility and schooling
decisions in germany and italy: the impact of secondary school tracks.
Checchi, D., et al. (2003). The italian educational system: family background
and social stratification. In Isae conference on monitoring italy.
Checchi, D., & Peragine, V. (2010). Inequality of opportunity in italy. The
Journal of Economic Inequality , 8 (4), 429–450.
Cooke, T. J., & Boyle, P. (2011). The migration of high school graduates to
college. Educational Evaluation and Policy Analysis , 33 (2), 202–213.
Dardanoni, V., Fields, G. S., Roemer, J. E., & Puerta, M. L. S. (2006).
How demanding should equality of opportunity be, and how much have we
achieved? Mobility and inequality: Frontiers of research from sociology and
economics , 59–82.
de Barros, R. P. (2009). Measuring inequality of opportunities in latin america
and the caribbean. World Bank Publications.
19
Dennett, A. (2012). Estimating flows between geographical locations:get me
started inspatial interaction modelling (Tech. Rep.). Citeseer.
De Vries, J. J., Nijkamp, P., & Rietveld, P. (2009). Exponential or power
distance-decay for commuting? an alternative specification. Environment
and Planning A, 41 (2), 461–480.
DiPrete, T. A., & Buchmann, C. (2006). Gender-specific trends in the value of
education and the emerging gender gap in college completion. Demography ,
43 (1), 1–24.
Dotti, N. F., Fratesi, U., Lenzi, C., & Percoco, M. (2013). Local labour
markets and the interregional mobility of italian university students. Spatial
Economic Analysis , 8 (4), 443–468.
Ferreira, F. H., & Gignoux, J. (2011). The measurement of inequality of
opportunity: Theory and an application to latin america. Review of Income
and Wealth, 57 (4), 622–657.
Flowerdew, R., & Aitkin, M. (1982). A method of fitting the gravity model
based on the poisson distribution*. Journal of regional science, 22 (2), 191–
202.
Foster, J. E., & Shneyerov, A. A. (2000). Path independent inequality mea-
sures. Journal of Economic Theory , 91 (2), 199–222.
Fotheringham, A., & O’Kelly, M. E. (1989). Spatial interaction models: for-
mulations and applications (Vol. 5). Kluwer Academic Pub.
Fotheringham, A. S., & Webber, M. J. (1980). Spatial structure and the
parameters of spatial interaction models. Geographical Analysis , 12 (1), 33–
46.
Frenette, M. (2003). Access to college and university: Does distance matter.
ottawa. Statistics Canada, Catalogue(11F0019MIE2003201).
20
Gibbons, S., & Vignoles, A. (2009). Access, choice and participation in higher
education.
Gitlesen, J. P., & Thorsen, I. (2000). A competing destinations approach
to modeling commuting flows: a theoretical interpretation and an empirical
application of the model. Environment and Planning A, 32 (11), 2057–2074.
Gossman, C. S., et al. (1967). Migration of college and university students in
the united states.
Hansen, W. G. (1959). How accessibility shapes land use. Journal of the
American Institute of planners , 25 (2), 73–76.
Haynes, K. E., & Fotheringham, A. S. (1984). Gravity and spatial interaction
models (Vol. 2). Sage publications Beverly Hills.
LeSage, J. P., & Pace, R. K. (2008). Spatial econometric modeling of origin-
destination flows*. Journal of Regional Science, 48 (5), 941–967.
Lowe, J. M., & Sen, A. (1996). Gravity model applications in health planning:
Analysis of an urban hospital market*. Journal of Regional Science, 36 (3),
437–461.
Lupi, C., & Ordine, P. (2009). Family income and students’mobility. Giornale
degli Economisti e Annali di Economia, 1–23.
Mayhew, L., Gibberd, R., & Hall, H. (1986). Predicting patient flows and
hospital case-mix. Environment and Planning A, 18 (5), 619–638.
McArthur, D. P., Kleppe, G., Thorsen, I., & Ubøe, J. (2011). The spatial
transferability of parameters in a gravity model of commuting flows. Journal
of Transport Geography , 19 (4), 596–605.
McHugh, R., & Morgan, J. N. (1984). The determinants of interstate student
migration: A place-to-place analysis. Economics of Education Review , 3 (4),
269–278.
21
Ordovensky, J. F. (1995). Effects of institutional attributes on enrollment
choice: Implications for postsecondary vocational education. Economics of
Education Review , 14 (4), 335–350.
Pigini, C., Staffolani, S., et al. (2013). Enrollment costs, university quality
and higher education choices in italy (Tech. Rep.). University Library of
Munich, Germany.
Reggiani, A., Bucci, P., & Russo, G. (2011). Accessibility and network struc-
tures in the german commuting. Networks and Spatial Economics , 11 (4),
621–641.
Roemer, J. E. (1998). Theories of distributive justice. Harvard University
Press.
Sa, C., Florax, R. J., & Rietveld, P. (2004). Determinants of the regional
demand for higher education in the netherlands: A gravity model approach.
Regional Studies , 38 (4), 375–392.
Sen, A., & Smith, T. (2012). Gravity models of spatial interaction behavior.
Springer Science & Business Media.
Singleton, A. D., Wilson, A., & OBrien, O. (2012). Geodemographics and
spatial interaction: an integrated model for higher education. Journal of
Geographical Systems , 14 (2), 223–241.
Smith, T. E. (1987). Poisson gravity models of spatial flows*. Journal of
Regional Science, 27 (3), 315–340.
Suhonen, T. (2014). Field-of-study choice in higher education: Does distance
matter? Spatial Economic Analysis , 9 (4), 355–375.
Talen, E., & Anselin, L. (1998). Assessing spatial equity: an evaluation of
measures of accessibility to public playgrounds. Environment and Planning
a, 30 (4), 595–613.
22
Tinto, V. (1973). College proximity and rates of college attendance. American
Educational Research Journal , 10 (4), 277–293.
Triventi, M., & Trivellato, P. (2009). Participation, performance and inequality
in italian higher education in the 20th century. Higher Education, 57 (6),
681–702.
Wilson, A. G. (1971). A family of spatial interaction models, and associated
developments. Environment and Planning A, 3 (1), 1–32.
Wilson, R. M., & Gibberd, R. (1990). Allocating resources between health
care regions: Catchment populations and allocating for equity. European
Journal of Operational Research, 49 (3), 333–347.
Yun, S., & Sen, A. (1994). Computation of maximum likelihood estimates of
gravity model parameters*. Journal of Regional Science, 34 (2), 199–216.
23
Table 1: Variables in analysis, data from ISTAT and MIUR
Variables Description
Residence Province of residence before enrollment (ISTAT)
Destination University Enrolled University 63 state and 14 private universities(ISTAT)
Distance Euclidean distance between city centroids and University addresses,
measured in QGIS based on coordinates
Sex 14,000 male and 17,400 female students graduated in 2007(ISTAT)
Parent’s Education The highest degree obtained by parents(ISTAT)
Two categories: at least one highly educated parent, both basic educated
“basic education” covers high school degree
high category at least bachelor’ s degree.
Financial Condition Occupation type of parents(ISTAT)
Three categories : both-high, one-high,both-low
High=Managers, Directors,High/Medium Qualification
Low=Office Worker, Lower-skilled workers
Sj Share of students who achieved highest scores (90-100)
from compulsory test before HE enrollment in the period 2002-2003 (MIUR)
Lj Proportion of limited places offered by universities to the total places (MIUR)
Ujl Institutional characteristics to be interacted with distance (MIUR)
Uj1 1 if private 0 otherwise
Uj2 1 if polytechnic 0 otherwise
Uj3 1 if south Uj4 1 if center Uj5 1 if island 0 otherwise
Types group1(both basic educated parents,male , low financial condition )
group2(both basic educated parents,female , low financial condition)
group3(both basic educated parents,male, medium financial condition)
group4(both basic educated parents, female, medium financial condition)
group5(both basic educated parents, male, high financial condition)
group6(both basic educated parents, female, high financial condition)
group7(at least one high educated parent, male, low financial condition )
group8(at least one high educated parent, female, low financial condition)
group9(at least one high educated parent, male, medium financial condition)
group10(at least one high educated parent, female, medium financial condition)
group11(at least one high educated parent, male, high financial condition)
group12(at least one high educated parent, female, high financial condition)
24
Table 2: Results of Poisson Regression First 5 Groups
Groups (1) (2) (3) (4) (7)
Variables Basic-Male-Lower Class Basic-Female-Lower Class Basic-Male-Middle Class Basic-Female-Middle Class ≥ 1high-Male-Lower Class
Sj 0.455 0.697 0.556 -0.087 1.502***
(0.421) (0.373) (0.595) (0.504) (0.396)
Lj 0.158 0.303*** 0.214 0.277* 0.140
(0.098) (0.086) (0.133) (0.119) (0.123)
µ 0.249*** 0.293*** 0.261*** 0.300*** 0.307***
(0.389) ( 0.032) (0.056) (0.045) (0.036)
Distance -1.535*** -1.709*** -1.530*** -1.574*** -1.625***
(0.047) ( 0.045) (0.069) (0.060) (0.046)
Institutional Interactions
x Private Univ. 0.726*** 0.608*** 0.509*** 0.599*** 0.631***
(0.070) ( 0.058) (0.093) (0.067) (0.057)
x Polytechnic 0.003 0.240 0.105 0.276* 0.134*
(0.072) (0.088) (0.105) (0.105) (0.022)
x South 0.561*** 0.359*** 0.549*** 0.386*** 432***
(0.062) (0.059) (0.090) (0.076) (0.065)
x Center 0.527*** 0.545*** 0.418*** 0.495*** 0.467***
(0.060) (0.058) (0.088) (0.078) (0.058)
x Island -0.206*** -0.039 -0.130 -0.375* -0.283*
(0.155) (0.122) (0.201) (0.173) (0.162)
(Intercept) 6.528*** 7.258*** 5.442*** 6.304*** 6.790***
(0.227) (0.202) (0.392) (0.291) (0.221)
Observations 1,572 1,572 1,572 1,572 1,572
R2 0.88 0.89 0.87 0.86 0.86
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Education of Parents-Gender of Student-Financial Condition of Parents
25
Table 3: Results of Poisson Regression Last 5 Groups
Groups (8) (9) (10) (11) (12)
Variables ≥1high-Lower Class ≥1high-Male-Middle Class ≥ 1high-Female-Middle Class ≥ 1high-Male-Higher Class ≥1high-Female-Higher Class
Sj 1.745*** 2.291*** 1.444*** 2.006*** 1.947***
(0.337) (0.313) (0.287) (0.293) (0.260)
Lj 0.384*** 0.184*** 0.187*** 0.112** 0.348***
(0.065) (0.073) (0.072) (0.082) (0.081)
µ 0.337*** 0.287*** 0.281*** 0.305*** 0.273***
(0.032) ( 0.032) (0.029) (0.029) (0.027)
Distance -1.581*** -1.551*** -1.532*** -1.522*** -1.499***
(0.041) ( 0.040) (0.034) ( 0.035) (0.032)
Institutional Interactions
x Private Univ. 0.590*** 0.490*** 0.481*** 0.549*** 0.573***
(0.045) ( 0.043) (0.038) (0.037) (0.033)
x Polytechnic -0.538 0.070 0.706 0.072 0.042
(0.078) (0.051) (0.061) (0.047) (0.054)
x South 0.123* 0.436*** 0.198*** 0.320*** 0.255***
(0.636) (0.609) (0.054) (0.054) (0.049)
x Center 0.399*** 0.499*** 0.418*** 0.384*** 0.407***
(0.053) (0.516) (0.045) (0.045) ( 0.040)
x Island -0.150 0.816 -0.252* -0.490*** -0.331**
(0.125) (0.130) (0.111) (0.118) (0.103)
(Intercept) 6.654*** 7.113*** 7.084*** 6.851*** 6.939***
(0.194) (0.187) (0.172) (0.170) (0.155)
Observations 1,572 1,572 1,572 1,572 1,572
R2 0.85 0.87 0.89 0.86 0.88
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Education of Parents-Gender of Student-Financial Condition of Parents
26
Table 4: Decomposition of Inequality In Access (MLD measures)
Spatial Inequality Inequality due to socioeconomic background Total Inequality
Inequality in Access to HE (First Approach) 0.35444 0.01776 0.37220
Inequality in Access to HE (Second Approach) 0.34637 0.02583 0.37220
Percentage Contribution (First Approach) %95 %5 %100
Percentage Contribution (Second Approach) %93 %7 %100
27
Figure 1: First 5 Groups
28
Figure 2: Last 5 Groups
29